Nearly all CFD methods can be considered as discretization methods for partial dierential equations, such as ยฎnite dierence, ยฎnite volume, ยฎnite element, spectral or boundary integral element methods. Virtually unrecognized by the scientiยฎc mainstream in computational ยฏuid dynamics (CFD) during the last decade, a completely dierent approach to ยฏow simulation has been developed in computational physics.
The basic idea of lattice-gas solvers (LGS) goes back to the cellular automation concept of John von Neumann. LGS use objects (cells'), being extremely simple compared to ยฎnite boxes or ยฎnite elements. The state of a cell is usually described by only a few bits therefore often two orders of magnitude more cells are used for a simulation with LGS than elements' in a ยฎnite element computation. LGS are explicit timestepping procedures; no equation systems have to be solved. Thus every time-step is extremely cheap in terms of CPU power compared to standard procedures, yet again much shorter time-steps have to be used. LGS are inherently parallel and are suitable to coarse-grain as well as to ยฎne-grain parallelization.
The paper will discuss some advantages and disadvantages of lattice-gas solvers and present LG simulation results of two-phase ยฏow with moving boundaries on a microscope scale for a two-dimensional test geometry of randomly distributed equally sized disks where the eect of surface tension on the steadystate saturation will be demonstrated.